Hardware-Accelerated Machine Learning (ML)-aided Electronic Design Automation (EDA) for Integrated Power Electronics Building Block (iPEBB)
Abstract
Objectives: The overarching objective is to drastically improve the efficiency of Machine Learning (ML)-based integrated power electronics building block (iPEBB) design by achieving a minimum 500x speed-up compared to current methods. We plan to accomplish this through three key strategies:- Utilizing FPGA accelerators to significantly speed up circuit simulations within a specialized Domain-Specific Architecture (DSA).- Developing accurate component modeling, including magnetics, and seamlessly applying it to the built simulation tool to ensure high-accuracy design results.- Applying Machine Learning to the Power Electronics Design Process for iPEBB with thorough validation.If successful, we expect to reduce the training time to mere minutes and implement a highly accurate designframework for iPEBB, thereby accelerating the rate of innovation in naval research.Technical Approaches:Four technical approaches have been devised to accomplish the objective outlined in the proposal: - FPGA Accelerators enhance circuit simulations with parallelism, pipelining, and techniques like selective time steps for speed.- Focus on accurately modeling critical components to improve the ML-based design framework s performance.- Develop and tailor an ML-based design framework for iPEBB, considering factors like parameters, tolerance, connectivity, and reliability.- An evaluation process and tool will validate and refine the design framework.Anticipated Outcome: The anticipated outcome of this project is a revolutionary advancement in the efficiency of ML-based iPEBB design,aiming to achieve a minimum 500x speed-up compared to current methods. It will be realized through multiple micro-outcomes from each task: - A modeling-based fast simulation with FPGA accelerators.- Accurate component modeling for iPEBB.- A robust ML-based designframework tailored to iPEBB.- A new standard and evaluation platform for assessing power electronics (iPEBB) design.Impact on DoD Capabilities: The outcome of this research-a rapid, ML-driven framework for PEBB design integrated with our newly developed high-speed simulation tool, offers transformative potential for a range of naval applications, including advanced propulsion systems, efficient energy storage, unmanned vehicles, shipboard power grids, etc. This groundbreaking work substantially boosts design productivity and the pace of innovation, a key element in the successful creation of future naval ships. Additionally, for the electrical circuitfor iPEBB, various models from different systems, such as mechanical systems and motors, will be applied to the built design framework based on the accelerator and ML-based design platform, which can improve the system-level modeling.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Nov 08, 2024
- Source ID
- N000142412633
Entities
People
- Yeonho Jeong
Organizations
- Office of Naval Research
- United States Navy
- University of Rhode Island